maemo.org - Talk

Quote:

Originally Posted by misiak (Post 1010328)

I also think that even the LXDE and OpenOffice timings show only 30(+/-10)% of speed improvement (see my previous post for calculations), they (mscion and karam) and us just use different measures... for me 100% speed improvement would be doing everything in (almost) 0 time... and doing something 2 times faster is 50%, not 100%... it's not even linear, god damn it... I think that proper formula is ((time_without_tweak-time_with_tweak)/time_without_tweak)*100%, not the strange one used by mscion. Let's imagine something took 10 seconds without overclocking.
1. Then after one tweak it takes 5 seconds.
Time gain = 5 seconds.
My way of counting = ((10-5)/10)*100% = 50% speed gain, so 1 second is 10 percentage points.
His way of counting = (10/5)*100%= 200% speed gain, so 1 second is 40 percentage points.
2. After removing this tweak and applying some different tweak it takes 3 seconds.
Time gain = 7 seconds.
My way of counting = 70% speed gain, so 1 second is 10 percentage points.
His way of counting = (10/3)*100% = 333% speed gain, so 1 second is 47 percentage points.
So it's not consistent...
and more important:
3. After removing the second tweak and applying some different, let's imagine device is less responsive and operation now takes 20 seconds.
Time gain = -10 seconds (10 seconds loss)
My way of counting = ((10-20)/10)*100% = -100% speed gain (100% speed loss), so 1 second is 10 percentage points.
His way of counting = (10/20)*100% = 50% speed gain, so 1 second is (50pp)/(-10)= -5 percentage points.

Summarizing:
- This strange way of counting speed gain used by mscion is inconsistent. Between situations 1 and 2 using your way of counting the speed gain jumps from 200 to 333 points suddenly, while using my way it's consistent (every second faster gives equal number of percentage points in speed gain %).
- This strange way of counting gives positive speed gain even if the speed gain is NEGATIVE, so it's definitely wrong.
- I think that speed gain is "seconds saved while doing some task", so it's (time_without_tweak - time_with_tweak). And to make it into percentage, one should take it, divide it by base time (time_without_tweak), so the formula is (time_without_tweak - time_with_tweak) / time_without_tweak. And multiplied by 100% ofc.

Thank you for reading ;) Maybe that's why in the thread title there is number "300%", it's actually around 60-70% speed gain (well, 2 times more than measured by these tests, but who knows, maybe on their devices something worked two times faster ;) it's more probable than 10 times faster ;) ) Anyone with me? :P

Ok, What have I done here! I was originally just following the convention of the thread's title. But, in reality, this convention is not correct mathematically. A percentage change should be given relative to the original value. For example if I earn $100 one week and then earn $105 the next week, my pay increased 5%. In terms of a formula it would be ((new_pay-original_pay)/original_pay)*100. If instead I had earned $95 that second week. Then my pay decreased 5%. So now getting back to the OCing tests. Assuming all things equal with karman and my N900,
The inital time without OCing is 35 seconds and the new time with OCing is 27 seconds. Then you can say the percentage change in time is ((27-35)/35)*100 = -23 % or that the time it took to launch the app decreased 23% with OCing. Note that I did not use the word "speed" here. I measured time.

Edit: Sorry if I got a little carried away... But... one additional thing, One can determine the percent improvement of speed by simply replacing the value of time by its reciprocal in the given formula (because speed scales as inverse time). For example, if launching an application took 10 seconds without OCing and then took 5 seconds with OCing then you write the percent increase in speed is ((1/5-1/10)/(1/10))*100= 100 %. So while one can say it runs twice as fast it is actually a 100 % improvement in speed as mentioned earlier.

In regards to the most favorable comparison of launching open office
Without OCing: 45
With OCing: 25 seconds

one can claim, from ((25-45)/45)*100, a 44 % decrease in launch time, a 80% increase in speed or a speed up of 1.8 times.